Prostate cancer radiomics: A study on IMRT response prediction based on MR image features and machine learning approaches

Introduction: To develop different radiomic models based on radiomic features and machine learning methods to predict early intensity modulated radiation therapy (IMRT) response.   Materials and Methods: Thirty prostate patients were included. All patients underwent pre ad post-IMRT T۲ weighted and apparent diffusing coefficient (ADC) magnetic resonance imaging (MRI). A wide range of radiomic features from different feature sets were extracted from all images. Delta radiomics was calculated as relative changes of pre-post-IMRT image features. Four feature selection methods and nine classification methods were evaluated in terms of their performance. We applied the ۵-fold cross-validation as the criterion for feature selection and classification. For IMRT response prediction, pre, post and Delta radiomic features were analyzed. Area under the curve (AUC) was calculated as model performance value. IMRT response was obtained by changes in ADC values . Results: For IMRT response prediction, ۱۵ models were developed. Pre-ADC model, unBalance/Select from Model/Adaptive Boosting had the highest predictive performance (AUC, ۰.۷۸). Conclusion:Radiomic models developed by MR Image features and machine learning approaches are noninvasive, easy and cost effective methods for personalized prostate cancer diagnosis and therapy.